On the Calculation of the l 2 → l 1 Induced Matrix Norm

We show that the l2 → l1 induced matrix norm, namely the norm induced by the l2 and l1 vector norms in the domain and range space, respectively, can be calculated as the maximal element of a finite set involving discrete additive combinations of the rows of the involved matrix with weights of ±1; the number of elements this set contains is exponential in the number of rows involved. A geometric interpretation of the result allows us to extend the result to some other induced norms. Finally, we generalize the findings to bounded linear operators on separable Banach spaces that can be obtained as strong limits of sequences of finite-dimensional linear operators.

[1]  Charles R. Johnson,et al.  Matrix analysis , 1985, Statistical Inference for Engineers and Data Scientists.

[2]  Andrea Montanari,et al.  The Noise-Sensitivity Phase Transition in Compressed Sensing , 2010, IEEE Transactions on Information Theory.

[3]  Emmanuel J. Candès,et al.  Near-Optimal Signal Recovery From Random Projections: Universal Encoding Strategies? , 2004, IEEE Transactions on Information Theory.

[4]  Michael S. Lewicki,et al.  Efficient auditory coding , 2006, Nature.

[5]  Bruno A Olshausen,et al.  Sparse coding of sensory inputs , 2004, Current Opinion in Neurobiology.

[6]  Michael S. Lewicki,et al.  Efficient coding of natural sounds , 2002, Nature Neuroscience.

[7]  D. Donoho,et al.  Maximal Sparsity Representation via l 1 Minimization , 2002 .

[8]  Barak A. Pearlmutter,et al.  Blind Source Separation by Sparse Decomposition in a Signal Dictionary , 2001, Neural Computation.

[9]  J. Rohn Computing the norm ∥A∥∞,1 is NP-hard , 2000 .

[10]  Terrence J. Sejnowski,et al.  Learning Overcomplete Representations , 2000, Neural Computation.

[11]  Michael A. Saunders,et al.  Atomic Decomposition by Basis Pursuit , 1998, SIAM J. Sci. Comput..

[12]  N. Higham Estimating the matrixp-norm , 1992 .